摘要
动态潜变量模型在多元状态监测中得到了广泛应用,从潜变量可预测性最大化出发,提出了一种基于预测特征分析(PFA)的状态监测方法。PFA采用无监督建模的形式,基于预测误差最小化计算全局最优可预测特征。进一步,构建T2和SPE统计量,对动态过程进行故障检测。以TE过程为对象,构建PFA模型,与典型的统计分析方法进行对比实验,结果表明,提出的方法能够准确高效地检测出故障,具有较高的可靠性和优越性。
The dynamic latent variable model has been widely used multivariate process monitoring.From the perspective of maximizing the predictability,this paper establishes a latent variable autoregressive model for process monitoring.The PFA algorithm adopts unsupervised model to extract optimal predictive features from the input signals,T2 and SPE statistics are further constructed to realize the monitoring of dynamic processes.Through experiments with typical methods,the proposed PFA based dynamic process monitoring method has superior performance and high reliability.
出处
《工业控制计算机》
2023年第7期5-6,9,共3页
Industrial Control Computer
关键词
动态过程
故障监测
主成分分析
预测特征分析
dynamic process
fault detection
principal component analysis
predictable feature analysis
